4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/f2fs_fs.h>
13 #include <linux/buffer_head.h>
14 #include <linux/mpage.h>
15 #include <linux/writeback.h>
16 #include <linux/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/prefetch.h>
21 #include <linux/uio.h>
23 #include <linux/memcontrol.h>
24 #include <linux/cleancache.h>
25 #include <linux/sched/signal.h>
31 #include <trace/events/f2fs.h>
33 static bool __is_cp_guaranteed(struct page *page)
35 struct address_space *mapping = page->mapping;
37 struct f2fs_sb_info *sbi;
42 inode = mapping->host;
43 sbi = F2FS_I_SB(inode);
45 if (inode->i_ino == F2FS_META_INO(sbi) ||
46 inode->i_ino == F2FS_NODE_INO(sbi) ||
47 S_ISDIR(inode->i_mode) ||
53 static void f2fs_read_end_io(struct bio *bio)
58 #ifdef CONFIG_F2FS_FAULT_INJECTION
59 if (time_to_inject(F2FS_P_SB(bio->bi_io_vec->bv_page), FAULT_IO)) {
60 f2fs_show_injection_info(FAULT_IO);
61 bio->bi_status = BLK_STS_IOERR;
65 if (f2fs_bio_encrypted(bio)) {
67 fscrypt_release_ctx(bio->bi_private);
69 fscrypt_decrypt_bio_pages(bio->bi_private, bio);
74 bio_for_each_segment_all(bvec, bio, i) {
75 struct page *page = bvec->bv_page;
77 if (!bio->bi_status) {
78 if (!PageUptodate(page))
79 SetPageUptodate(page);
81 ClearPageUptodate(page);
89 static void f2fs_write_end_io(struct bio *bio)
91 struct f2fs_sb_info *sbi = bio->bi_private;
95 bio_for_each_segment_all(bvec, bio, i) {
96 struct page *page = bvec->bv_page;
97 enum count_type type = WB_DATA_TYPE(page);
99 if (IS_DUMMY_WRITTEN_PAGE(page)) {
100 set_page_private(page, (unsigned long)NULL);
101 ClearPagePrivate(page);
103 mempool_free(page, sbi->write_io_dummy);
105 if (unlikely(bio->bi_status))
106 f2fs_stop_checkpoint(sbi, true);
110 fscrypt_pullback_bio_page(&page, true);
112 if (unlikely(bio->bi_status)) {
113 mapping_set_error(page->mapping, -EIO);
114 f2fs_stop_checkpoint(sbi, true);
116 dec_page_count(sbi, type);
117 clear_cold_data(page);
118 end_page_writeback(page);
120 if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
121 wq_has_sleeper(&sbi->cp_wait))
122 wake_up(&sbi->cp_wait);
128 * Return true, if pre_bio's bdev is same as its target device.
130 struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
131 block_t blk_addr, struct bio *bio)
133 struct block_device *bdev = sbi->sb->s_bdev;
136 for (i = 0; i < sbi->s_ndevs; i++) {
137 if (FDEV(i).start_blk <= blk_addr &&
138 FDEV(i).end_blk >= blk_addr) {
139 blk_addr -= FDEV(i).start_blk;
146 bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(blk_addr);
151 int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
155 for (i = 0; i < sbi->s_ndevs; i++)
156 if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
161 static bool __same_bdev(struct f2fs_sb_info *sbi,
162 block_t blk_addr, struct bio *bio)
164 return f2fs_target_device(sbi, blk_addr, NULL) == bio->bi_bdev;
168 * Low-level block read/write IO operations.
170 static struct bio *__bio_alloc(struct f2fs_sb_info *sbi, block_t blk_addr,
171 int npages, bool is_read)
175 bio = f2fs_bio_alloc(npages);
177 f2fs_target_device(sbi, blk_addr, bio);
178 bio->bi_end_io = is_read ? f2fs_read_end_io : f2fs_write_end_io;
179 bio->bi_private = is_read ? NULL : sbi;
184 static inline void __submit_bio(struct f2fs_sb_info *sbi,
185 struct bio *bio, enum page_type type)
187 if (!is_read_io(bio_op(bio))) {
190 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
191 current->plug && (type == DATA || type == NODE))
192 blk_finish_plug(current->plug);
194 if (type != DATA && type != NODE)
197 start = bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS;
198 start %= F2FS_IO_SIZE(sbi);
203 /* fill dummy pages */
204 for (; start < F2FS_IO_SIZE(sbi); start++) {
206 mempool_alloc(sbi->write_io_dummy,
207 GFP_NOIO | __GFP_ZERO | __GFP_NOFAIL);
208 f2fs_bug_on(sbi, !page);
210 SetPagePrivate(page);
211 set_page_private(page, (unsigned long)DUMMY_WRITTEN_PAGE);
213 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
217 * In the NODE case, we lose next block address chain. So, we
218 * need to do checkpoint in f2fs_sync_file.
221 set_sbi_flag(sbi, SBI_NEED_CP);
224 if (is_read_io(bio_op(bio)))
225 trace_f2fs_submit_read_bio(sbi->sb, type, bio);
227 trace_f2fs_submit_write_bio(sbi->sb, type, bio);
231 static void __submit_merged_bio(struct f2fs_bio_info *io)
233 struct f2fs_io_info *fio = &io->fio;
238 bio_set_op_attrs(io->bio, fio->op, fio->op_flags);
240 if (is_read_io(fio->op))
241 trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
243 trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
245 __submit_bio(io->sbi, io->bio, fio->type);
249 static bool __has_merged_page(struct f2fs_bio_info *io,
250 struct inode *inode, nid_t ino, pgoff_t idx)
252 struct bio_vec *bvec;
262 bio_for_each_segment_all(bvec, io->bio, i) {
264 if (bvec->bv_page->mapping)
265 target = bvec->bv_page;
267 target = fscrypt_control_page(bvec->bv_page);
269 if (idx != target->index)
272 if (inode && inode == target->mapping->host)
274 if (ino && ino == ino_of_node(target))
281 static bool has_merged_page(struct f2fs_sb_info *sbi, struct inode *inode,
282 nid_t ino, pgoff_t idx, enum page_type type)
284 enum page_type btype = PAGE_TYPE_OF_BIO(type);
286 struct f2fs_bio_info *io;
289 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
290 io = sbi->write_io[btype] + temp;
292 down_read(&io->io_rwsem);
293 ret = __has_merged_page(io, inode, ino, idx);
294 up_read(&io->io_rwsem);
296 /* TODO: use HOT temp only for meta pages now. */
297 if (ret || btype == META)
303 static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
304 enum page_type type, enum temp_type temp)
306 enum page_type btype = PAGE_TYPE_OF_BIO(type);
307 struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
309 down_write(&io->io_rwsem);
311 /* change META to META_FLUSH in the checkpoint procedure */
312 if (type >= META_FLUSH) {
313 io->fio.type = META_FLUSH;
314 io->fio.op = REQ_OP_WRITE;
315 io->fio.op_flags = REQ_META | REQ_PRIO | REQ_SYNC;
316 if (!test_opt(sbi, NOBARRIER))
317 io->fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
319 __submit_merged_bio(io);
320 up_write(&io->io_rwsem);
323 static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
324 struct inode *inode, nid_t ino, pgoff_t idx,
325 enum page_type type, bool force)
329 if (!force && !has_merged_page(sbi, inode, ino, idx, type))
332 for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
334 __f2fs_submit_merged_write(sbi, type, temp);
336 /* TODO: use HOT temp only for meta pages now. */
342 void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
344 __submit_merged_write_cond(sbi, NULL, 0, 0, type, true);
347 void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
348 struct inode *inode, nid_t ino, pgoff_t idx,
351 __submit_merged_write_cond(sbi, inode, ino, idx, type, false);
354 void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
356 f2fs_submit_merged_write(sbi, DATA);
357 f2fs_submit_merged_write(sbi, NODE);
358 f2fs_submit_merged_write(sbi, META);
362 * Fill the locked page with data located in the block address.
363 * A caller needs to unlock the page on failure.
365 int f2fs_submit_page_bio(struct f2fs_io_info *fio)
368 struct page *page = fio->encrypted_page ?
369 fio->encrypted_page : fio->page;
371 trace_f2fs_submit_page_bio(page, fio);
372 f2fs_trace_ios(fio, 0);
374 /* Allocate a new bio */
375 bio = __bio_alloc(fio->sbi, fio->new_blkaddr, 1, is_read_io(fio->op));
377 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
381 bio_set_op_attrs(bio, fio->op, fio->op_flags);
383 __submit_bio(fio->sbi, bio, fio->type);
385 if (!is_read_io(fio->op))
386 inc_page_count(fio->sbi, WB_DATA_TYPE(fio->page));
390 int f2fs_submit_page_write(struct f2fs_io_info *fio)
392 struct f2fs_sb_info *sbi = fio->sbi;
393 enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
394 struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
395 struct page *bio_page;
398 f2fs_bug_on(sbi, is_read_io(fio->op));
400 down_write(&io->io_rwsem);
403 spin_lock(&io->io_lock);
404 if (list_empty(&io->io_list)) {
405 spin_unlock(&io->io_lock);
408 fio = list_first_entry(&io->io_list,
409 struct f2fs_io_info, list);
410 list_del(&fio->list);
411 spin_unlock(&io->io_lock);
414 if (fio->old_blkaddr != NEW_ADDR)
415 verify_block_addr(sbi, fio->old_blkaddr);
416 verify_block_addr(sbi, fio->new_blkaddr);
418 bio_page = fio->encrypted_page ? fio->encrypted_page : fio->page;
420 /* set submitted = 1 as a return value */
423 inc_page_count(sbi, WB_DATA_TYPE(bio_page));
425 if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 ||
426 (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) ||
427 !__same_bdev(sbi, fio->new_blkaddr, io->bio)))
428 __submit_merged_bio(io);
430 if (io->bio == NULL) {
431 if ((fio->type == DATA || fio->type == NODE) &&
432 fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
434 dec_page_count(sbi, WB_DATA_TYPE(bio_page));
437 io->bio = __bio_alloc(sbi, fio->new_blkaddr,
438 BIO_MAX_PAGES, false);
442 if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
443 __submit_merged_bio(io);
447 io->last_block_in_bio = fio->new_blkaddr;
448 f2fs_trace_ios(fio, 0);
450 trace_f2fs_submit_page_write(fio->page, fio);
455 up_write(&io->io_rwsem);
459 static void __set_data_blkaddr(struct dnode_of_data *dn)
461 struct f2fs_node *rn = F2FS_NODE(dn->node_page);
465 if (IS_INODE(dn->node_page) && f2fs_has_extra_attr(dn->inode))
466 base = get_extra_isize(dn->inode);
468 /* Get physical address of data block */
469 addr_array = blkaddr_in_node(rn);
470 addr_array[base + dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
474 * Lock ordering for the change of data block address:
477 * update block addresses in the node page
479 void set_data_blkaddr(struct dnode_of_data *dn)
481 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
482 __set_data_blkaddr(dn);
483 if (set_page_dirty(dn->node_page))
484 dn->node_changed = true;
487 void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
489 dn->data_blkaddr = blkaddr;
490 set_data_blkaddr(dn);
491 f2fs_update_extent_cache(dn);
494 /* dn->ofs_in_node will be returned with up-to-date last block pointer */
495 int reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
497 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
503 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
505 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
508 trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
509 dn->ofs_in_node, count);
511 f2fs_wait_on_page_writeback(dn->node_page, NODE, true);
513 for (; count > 0; dn->ofs_in_node++) {
514 block_t blkaddr = datablock_addr(dn->inode,
515 dn->node_page, dn->ofs_in_node);
516 if (blkaddr == NULL_ADDR) {
517 dn->data_blkaddr = NEW_ADDR;
518 __set_data_blkaddr(dn);
523 if (set_page_dirty(dn->node_page))
524 dn->node_changed = true;
528 /* Should keep dn->ofs_in_node unchanged */
529 int reserve_new_block(struct dnode_of_data *dn)
531 unsigned int ofs_in_node = dn->ofs_in_node;
534 ret = reserve_new_blocks(dn, 1);
535 dn->ofs_in_node = ofs_in_node;
539 int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
541 bool need_put = dn->inode_page ? false : true;
544 err = get_dnode_of_data(dn, index, ALLOC_NODE);
548 if (dn->data_blkaddr == NULL_ADDR)
549 err = reserve_new_block(dn);
555 int f2fs_get_block(struct dnode_of_data *dn, pgoff_t index)
557 struct extent_info ei = {0,0,0};
558 struct inode *inode = dn->inode;
560 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
561 dn->data_blkaddr = ei.blk + index - ei.fofs;
565 return f2fs_reserve_block(dn, index);
568 struct page *get_read_data_page(struct inode *inode, pgoff_t index,
569 int op_flags, bool for_write)
571 struct address_space *mapping = inode->i_mapping;
572 struct dnode_of_data dn;
574 struct extent_info ei = {0,0,0};
576 struct f2fs_io_info fio = {
577 .sbi = F2FS_I_SB(inode),
580 .op_flags = op_flags,
581 .encrypted_page = NULL,
584 if (f2fs_encrypted_file(inode))
585 return read_mapping_page(mapping, index, NULL);
587 page = f2fs_grab_cache_page(mapping, index, for_write);
589 return ERR_PTR(-ENOMEM);
591 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
592 dn.data_blkaddr = ei.blk + index - ei.fofs;
596 set_new_dnode(&dn, inode, NULL, NULL, 0);
597 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
602 if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
607 if (PageUptodate(page)) {
613 * A new dentry page is allocated but not able to be written, since its
614 * new inode page couldn't be allocated due to -ENOSPC.
615 * In such the case, its blkaddr can be remained as NEW_ADDR.
616 * see, f2fs_add_link -> get_new_data_page -> init_inode_metadata.
618 if (dn.data_blkaddr == NEW_ADDR) {
619 zero_user_segment(page, 0, PAGE_SIZE);
620 if (!PageUptodate(page))
621 SetPageUptodate(page);
626 fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
628 err = f2fs_submit_page_bio(&fio);
634 f2fs_put_page(page, 1);
638 struct page *find_data_page(struct inode *inode, pgoff_t index)
640 struct address_space *mapping = inode->i_mapping;
643 page = find_get_page(mapping, index);
644 if (page && PageUptodate(page))
646 f2fs_put_page(page, 0);
648 page = get_read_data_page(inode, index, 0, false);
652 if (PageUptodate(page))
655 wait_on_page_locked(page);
656 if (unlikely(!PageUptodate(page))) {
657 f2fs_put_page(page, 0);
658 return ERR_PTR(-EIO);
664 * If it tries to access a hole, return an error.
665 * Because, the callers, functions in dir.c and GC, should be able to know
666 * whether this page exists or not.
668 struct page *get_lock_data_page(struct inode *inode, pgoff_t index,
671 struct address_space *mapping = inode->i_mapping;
674 page = get_read_data_page(inode, index, 0, for_write);
678 /* wait for read completion */
680 if (unlikely(page->mapping != mapping)) {
681 f2fs_put_page(page, 1);
684 if (unlikely(!PageUptodate(page))) {
685 f2fs_put_page(page, 1);
686 return ERR_PTR(-EIO);
692 * Caller ensures that this data page is never allocated.
693 * A new zero-filled data page is allocated in the page cache.
695 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
697 * Note that, ipage is set only by make_empty_dir, and if any error occur,
698 * ipage should be released by this function.
700 struct page *get_new_data_page(struct inode *inode,
701 struct page *ipage, pgoff_t index, bool new_i_size)
703 struct address_space *mapping = inode->i_mapping;
705 struct dnode_of_data dn;
708 page = f2fs_grab_cache_page(mapping, index, true);
711 * before exiting, we should make sure ipage will be released
712 * if any error occur.
714 f2fs_put_page(ipage, 1);
715 return ERR_PTR(-ENOMEM);
718 set_new_dnode(&dn, inode, ipage, NULL, 0);
719 err = f2fs_reserve_block(&dn, index);
721 f2fs_put_page(page, 1);
727 if (PageUptodate(page))
730 if (dn.data_blkaddr == NEW_ADDR) {
731 zero_user_segment(page, 0, PAGE_SIZE);
732 if (!PageUptodate(page))
733 SetPageUptodate(page);
735 f2fs_put_page(page, 1);
737 /* if ipage exists, blkaddr should be NEW_ADDR */
738 f2fs_bug_on(F2FS_I_SB(inode), ipage);
739 page = get_lock_data_page(inode, index, true);
744 if (new_i_size && i_size_read(inode) <
745 ((loff_t)(index + 1) << PAGE_SHIFT))
746 f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
750 static int __allocate_data_block(struct dnode_of_data *dn)
752 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
753 struct f2fs_summary sum;
759 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
762 dn->data_blkaddr = datablock_addr(dn->inode,
763 dn->node_page, dn->ofs_in_node);
764 if (dn->data_blkaddr == NEW_ADDR)
767 if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
771 get_node_info(sbi, dn->nid, &ni);
772 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
774 allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr,
775 &sum, CURSEG_WARM_DATA, NULL, false);
776 set_data_blkaddr(dn);
779 fofs = start_bidx_of_node(ofs_of_node(dn->node_page), dn->inode) +
781 if (i_size_read(dn->inode) < ((loff_t)(fofs + 1) << PAGE_SHIFT))
782 f2fs_i_size_write(dn->inode,
783 ((loff_t)(fofs + 1) << PAGE_SHIFT));
787 static inline bool __force_buffered_io(struct inode *inode, int rw)
789 return (f2fs_encrypted_file(inode) ||
790 (rw == WRITE && test_opt(F2FS_I_SB(inode), LFS)) ||
791 F2FS_I_SB(inode)->s_ndevs);
794 int f2fs_preallocate_blocks(struct kiocb *iocb, struct iov_iter *from)
796 struct inode *inode = file_inode(iocb->ki_filp);
797 struct f2fs_map_blocks map;
800 if (is_inode_flag_set(inode, FI_NO_PREALLOC))
803 map.m_lblk = F2FS_BLK_ALIGN(iocb->ki_pos);
804 map.m_len = F2FS_BYTES_TO_BLK(iocb->ki_pos + iov_iter_count(from));
805 if (map.m_len > map.m_lblk)
806 map.m_len -= map.m_lblk;
810 map.m_next_pgofs = NULL;
812 if (iocb->ki_flags & IOCB_DIRECT) {
813 err = f2fs_convert_inline_inode(inode);
816 return f2fs_map_blocks(inode, &map, 1,
817 __force_buffered_io(inode, WRITE) ?
818 F2FS_GET_BLOCK_PRE_AIO :
819 F2FS_GET_BLOCK_PRE_DIO);
821 if (iocb->ki_pos + iov_iter_count(from) > MAX_INLINE_DATA(inode)) {
822 err = f2fs_convert_inline_inode(inode);
826 if (!f2fs_has_inline_data(inode))
827 return f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
831 static inline void __do_map_lock(struct f2fs_sb_info *sbi, int flag, bool lock)
833 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
835 down_read(&sbi->node_change);
837 up_read(&sbi->node_change);
847 * f2fs_map_blocks() now supported readahead/bmap/rw direct_IO with
848 * f2fs_map_blocks structure.
849 * If original data blocks are allocated, then give them to blockdev.
851 * a. preallocate requested block addresses
852 * b. do not use extent cache for better performance
853 * c. give the block addresses to blockdev
855 int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map,
856 int create, int flag)
858 unsigned int maxblocks = map->m_len;
859 struct dnode_of_data dn;
860 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
861 int mode = create ? ALLOC_NODE : LOOKUP_NODE;
862 pgoff_t pgofs, end_offset, end;
863 int err = 0, ofs = 1;
864 unsigned int ofs_in_node, last_ofs_in_node;
866 struct extent_info ei = {0,0,0};
875 /* it only supports block size == page size */
876 pgofs = (pgoff_t)map->m_lblk;
877 end = pgofs + maxblocks;
879 if (!create && f2fs_lookup_extent_cache(inode, pgofs, &ei)) {
880 map->m_pblk = ei.blk + pgofs - ei.fofs;
881 map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgofs);
882 map->m_flags = F2FS_MAP_MAPPED;
888 __do_map_lock(sbi, flag, true);
890 /* When reading holes, we need its node page */
891 set_new_dnode(&dn, inode, NULL, NULL, 0);
892 err = get_dnode_of_data(&dn, pgofs, mode);
894 if (flag == F2FS_GET_BLOCK_BMAP)
896 if (err == -ENOENT) {
898 if (map->m_next_pgofs)
900 get_next_page_offset(&dn, pgofs);
906 last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
907 end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
910 blkaddr = datablock_addr(dn.inode, dn.node_page, dn.ofs_in_node);
912 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR) {
914 if (unlikely(f2fs_cp_error(sbi))) {
918 if (flag == F2FS_GET_BLOCK_PRE_AIO) {
919 if (blkaddr == NULL_ADDR) {
921 last_ofs_in_node = dn.ofs_in_node;
924 err = __allocate_data_block(&dn);
926 set_inode_flag(inode, FI_APPEND_WRITE);
930 map->m_flags |= F2FS_MAP_NEW;
931 blkaddr = dn.data_blkaddr;
933 if (flag == F2FS_GET_BLOCK_BMAP) {
937 if (flag == F2FS_GET_BLOCK_FIEMAP &&
938 blkaddr == NULL_ADDR) {
939 if (map->m_next_pgofs)
940 *map->m_next_pgofs = pgofs + 1;
942 if (flag != F2FS_GET_BLOCK_FIEMAP ||
948 if (flag == F2FS_GET_BLOCK_PRE_AIO)
951 if (map->m_len == 0) {
952 /* preallocated unwritten block should be mapped for fiemap. */
953 if (blkaddr == NEW_ADDR)
954 map->m_flags |= F2FS_MAP_UNWRITTEN;
955 map->m_flags |= F2FS_MAP_MAPPED;
957 map->m_pblk = blkaddr;
959 } else if ((map->m_pblk != NEW_ADDR &&
960 blkaddr == (map->m_pblk + ofs)) ||
961 (map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
962 flag == F2FS_GET_BLOCK_PRE_DIO) {
973 /* preallocate blocks in batch for one dnode page */
974 if (flag == F2FS_GET_BLOCK_PRE_AIO &&
975 (pgofs == end || dn.ofs_in_node == end_offset)) {
977 dn.ofs_in_node = ofs_in_node;
978 err = reserve_new_blocks(&dn, prealloc);
982 map->m_len += dn.ofs_in_node - ofs_in_node;
983 if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
987 dn.ofs_in_node = end_offset;
992 else if (dn.ofs_in_node < end_offset)
998 __do_map_lock(sbi, flag, false);
999 f2fs_balance_fs(sbi, dn.node_changed);
1004 f2fs_put_dnode(&dn);
1007 __do_map_lock(sbi, flag, false);
1008 f2fs_balance_fs(sbi, dn.node_changed);
1011 trace_f2fs_map_blocks(inode, map, err);
1015 static int __get_data_block(struct inode *inode, sector_t iblock,
1016 struct buffer_head *bh, int create, int flag,
1017 pgoff_t *next_pgofs)
1019 struct f2fs_map_blocks map;
1022 map.m_lblk = iblock;
1023 map.m_len = bh->b_size >> inode->i_blkbits;
1024 map.m_next_pgofs = next_pgofs;
1026 err = f2fs_map_blocks(inode, &map, create, flag);
1028 map_bh(bh, inode->i_sb, map.m_pblk);
1029 bh->b_state = (bh->b_state & ~F2FS_MAP_FLAGS) | map.m_flags;
1030 bh->b_size = (u64)map.m_len << inode->i_blkbits;
1035 static int get_data_block(struct inode *inode, sector_t iblock,
1036 struct buffer_head *bh_result, int create, int flag,
1037 pgoff_t *next_pgofs)
1039 return __get_data_block(inode, iblock, bh_result, create,
1043 static int get_data_block_dio(struct inode *inode, sector_t iblock,
1044 struct buffer_head *bh_result, int create)
1046 return __get_data_block(inode, iblock, bh_result, create,
1047 F2FS_GET_BLOCK_DEFAULT, NULL);
1050 static int get_data_block_bmap(struct inode *inode, sector_t iblock,
1051 struct buffer_head *bh_result, int create)
1053 /* Block number less than F2FS MAX BLOCKS */
1054 if (unlikely(iblock >= F2FS_I_SB(inode)->max_file_blocks))
1057 return __get_data_block(inode, iblock, bh_result, create,
1058 F2FS_GET_BLOCK_BMAP, NULL);
1061 static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
1063 return (offset >> inode->i_blkbits);
1066 static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
1068 return (blk << inode->i_blkbits);
1071 int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1074 struct buffer_head map_bh;
1075 sector_t start_blk, last_blk;
1077 u64 logical = 0, phys = 0, size = 0;
1081 ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
1085 if (f2fs_has_inline_data(inode)) {
1086 ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
1093 if (logical_to_blk(inode, len) == 0)
1094 len = blk_to_logical(inode, 1);
1096 start_blk = logical_to_blk(inode, start);
1097 last_blk = logical_to_blk(inode, start + len - 1);
1100 memset(&map_bh, 0, sizeof(struct buffer_head));
1101 map_bh.b_size = len;
1103 ret = get_data_block(inode, start_blk, &map_bh, 0,
1104 F2FS_GET_BLOCK_FIEMAP, &next_pgofs);
1109 if (!buffer_mapped(&map_bh)) {
1110 start_blk = next_pgofs;
1112 if (blk_to_logical(inode, start_blk) < blk_to_logical(inode,
1113 F2FS_I_SB(inode)->max_file_blocks))
1116 flags |= FIEMAP_EXTENT_LAST;
1120 if (f2fs_encrypted_inode(inode))
1121 flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
1123 ret = fiemap_fill_next_extent(fieinfo, logical,
1127 if (start_blk > last_blk || ret)
1130 logical = blk_to_logical(inode, start_blk);
1131 phys = blk_to_logical(inode, map_bh.b_blocknr);
1132 size = map_bh.b_size;
1134 if (buffer_unwritten(&map_bh))
1135 flags = FIEMAP_EXTENT_UNWRITTEN;
1137 start_blk += logical_to_blk(inode, size);
1141 if (fatal_signal_pending(current))
1149 inode_unlock(inode);
1153 static struct bio *f2fs_grab_bio(struct inode *inode, block_t blkaddr,
1156 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1157 struct fscrypt_ctx *ctx = NULL;
1160 if (f2fs_encrypted_file(inode)) {
1161 ctx = fscrypt_get_ctx(inode, GFP_NOFS);
1163 return ERR_CAST(ctx);
1165 /* wait the page to be moved by cleaning */
1166 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1169 bio = bio_alloc(GFP_KERNEL, min_t(int, nr_pages, BIO_MAX_PAGES));
1172 fscrypt_release_ctx(ctx);
1173 return ERR_PTR(-ENOMEM);
1175 f2fs_target_device(sbi, blkaddr, bio);
1176 bio->bi_end_io = f2fs_read_end_io;
1177 bio->bi_private = ctx;
1183 * This function was originally taken from fs/mpage.c, and customized for f2fs.
1184 * Major change was from block_size == page_size in f2fs by default.
1186 static int f2fs_mpage_readpages(struct address_space *mapping,
1187 struct list_head *pages, struct page *page,
1190 struct bio *bio = NULL;
1192 sector_t last_block_in_bio = 0;
1193 struct inode *inode = mapping->host;
1194 const unsigned blkbits = inode->i_blkbits;
1195 const unsigned blocksize = 1 << blkbits;
1196 sector_t block_in_file;
1197 sector_t last_block;
1198 sector_t last_block_in_file;
1200 struct f2fs_map_blocks map;
1206 map.m_next_pgofs = NULL;
1208 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
1211 page = list_last_entry(pages, struct page, lru);
1213 prefetchw(&page->flags);
1214 list_del(&page->lru);
1215 if (add_to_page_cache_lru(page, mapping,
1217 readahead_gfp_mask(mapping)))
1221 block_in_file = (sector_t)page->index;
1222 last_block = block_in_file + nr_pages;
1223 last_block_in_file = (i_size_read(inode) + blocksize - 1) >>
1225 if (last_block > last_block_in_file)
1226 last_block = last_block_in_file;
1229 * Map blocks using the previous result first.
1231 if ((map.m_flags & F2FS_MAP_MAPPED) &&
1232 block_in_file > map.m_lblk &&
1233 block_in_file < (map.m_lblk + map.m_len))
1237 * Then do more f2fs_map_blocks() calls until we are
1238 * done with this page.
1242 if (block_in_file < last_block) {
1243 map.m_lblk = block_in_file;
1244 map.m_len = last_block - block_in_file;
1246 if (f2fs_map_blocks(inode, &map, 0,
1247 F2FS_GET_BLOCK_DEFAULT))
1248 goto set_error_page;
1251 if ((map.m_flags & F2FS_MAP_MAPPED)) {
1252 block_nr = map.m_pblk + block_in_file - map.m_lblk;
1253 SetPageMappedToDisk(page);
1255 if (!PageUptodate(page) && !cleancache_get_page(page)) {
1256 SetPageUptodate(page);
1260 zero_user_segment(page, 0, PAGE_SIZE);
1261 if (!PageUptodate(page))
1262 SetPageUptodate(page);
1268 * This page will go to BIO. Do we need to send this
1271 if (bio && (last_block_in_bio != block_nr - 1 ||
1272 !__same_bdev(F2FS_I_SB(inode), block_nr, bio))) {
1274 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1278 bio = f2fs_grab_bio(inode, block_nr, nr_pages);
1281 goto set_error_page;
1283 bio_set_op_attrs(bio, REQ_OP_READ, 0);
1286 if (bio_add_page(bio, page, blocksize, 0) < blocksize)
1287 goto submit_and_realloc;
1289 last_block_in_bio = block_nr;
1293 zero_user_segment(page, 0, PAGE_SIZE);
1298 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1306 BUG_ON(pages && !list_empty(pages));
1308 __submit_bio(F2FS_I_SB(inode), bio, DATA);
1312 static int f2fs_read_data_page(struct file *file, struct page *page)
1314 struct inode *inode = page->mapping->host;
1317 trace_f2fs_readpage(page, DATA);
1319 /* If the file has inline data, try to read it directly */
1320 if (f2fs_has_inline_data(inode))
1321 ret = f2fs_read_inline_data(inode, page);
1323 ret = f2fs_mpage_readpages(page->mapping, NULL, page, 1);
1327 static int f2fs_read_data_pages(struct file *file,
1328 struct address_space *mapping,
1329 struct list_head *pages, unsigned nr_pages)
1331 struct inode *inode = file->f_mapping->host;
1332 struct page *page = list_last_entry(pages, struct page, lru);
1334 trace_f2fs_readpages(inode, page, nr_pages);
1336 /* If the file has inline data, skip readpages */
1337 if (f2fs_has_inline_data(inode))
1340 return f2fs_mpage_readpages(mapping, pages, NULL, nr_pages);
1343 static int encrypt_one_page(struct f2fs_io_info *fio)
1345 struct inode *inode = fio->page->mapping->host;
1346 gfp_t gfp_flags = GFP_NOFS;
1348 if (!f2fs_encrypted_file(inode))
1351 /* wait for GCed encrypted page writeback */
1352 f2fs_wait_on_encrypted_page_writeback(fio->sbi, fio->old_blkaddr);
1355 fio->encrypted_page = fscrypt_encrypt_page(inode, fio->page,
1356 PAGE_SIZE, 0, fio->page->index, gfp_flags);
1357 if (!IS_ERR(fio->encrypted_page))
1360 /* flush pending IOs and wait for a while in the ENOMEM case */
1361 if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
1362 f2fs_flush_merged_writes(fio->sbi);
1363 congestion_wait(BLK_RW_ASYNC, HZ/50);
1364 gfp_flags |= __GFP_NOFAIL;
1367 return PTR_ERR(fio->encrypted_page);
1370 static inline bool need_inplace_update(struct f2fs_io_info *fio)
1372 struct inode *inode = fio->page->mapping->host;
1374 if (S_ISDIR(inode->i_mode) || f2fs_is_atomic_file(inode))
1376 if (is_cold_data(fio->page))
1378 if (IS_ATOMIC_WRITTEN_PAGE(fio->page))
1381 return need_inplace_update_policy(inode, fio);
1384 static inline bool valid_ipu_blkaddr(struct f2fs_io_info *fio)
1386 if (fio->old_blkaddr == NEW_ADDR)
1388 if (fio->old_blkaddr == NULL_ADDR)
1393 int do_write_data_page(struct f2fs_io_info *fio)
1395 struct page *page = fio->page;
1396 struct inode *inode = page->mapping->host;
1397 struct dnode_of_data dn;
1398 struct extent_info ei = {0,0,0};
1399 bool ipu_force = false;
1402 set_new_dnode(&dn, inode, NULL, NULL, 0);
1403 if (need_inplace_update(fio) &&
1404 f2fs_lookup_extent_cache(inode, page->index, &ei)) {
1405 fio->old_blkaddr = ei.blk + page->index - ei.fofs;
1407 if (valid_ipu_blkaddr(fio)) {
1409 fio->need_lock = LOCK_DONE;
1414 /* Deadlock due to between page->lock and f2fs_lock_op */
1415 if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
1418 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
1422 fio->old_blkaddr = dn.data_blkaddr;
1424 /* This page is already truncated */
1425 if (fio->old_blkaddr == NULL_ADDR) {
1426 ClearPageUptodate(page);
1431 * If current allocation needs SSR,
1432 * it had better in-place writes for updated data.
1434 if (ipu_force || (valid_ipu_blkaddr(fio) && need_inplace_update(fio))) {
1435 err = encrypt_one_page(fio);
1439 set_page_writeback(page);
1440 f2fs_put_dnode(&dn);
1441 if (fio->need_lock == LOCK_REQ)
1442 f2fs_unlock_op(fio->sbi);
1443 err = rewrite_data_page(fio);
1444 trace_f2fs_do_write_data_page(fio->page, IPU);
1445 set_inode_flag(inode, FI_UPDATE_WRITE);
1449 if (fio->need_lock == LOCK_RETRY) {
1450 if (!f2fs_trylock_op(fio->sbi)) {
1454 fio->need_lock = LOCK_REQ;
1457 err = encrypt_one_page(fio);
1461 set_page_writeback(page);
1463 /* LFS mode write path */
1464 write_data_page(&dn, fio);
1465 trace_f2fs_do_write_data_page(page, OPU);
1466 set_inode_flag(inode, FI_APPEND_WRITE);
1467 if (page->index == 0)
1468 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
1470 f2fs_put_dnode(&dn);
1472 if (fio->need_lock == LOCK_REQ)
1473 f2fs_unlock_op(fio->sbi);
1477 static int __write_data_page(struct page *page, bool *submitted,
1478 struct writeback_control *wbc,
1479 enum iostat_type io_type)
1481 struct inode *inode = page->mapping->host;
1482 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1483 loff_t i_size = i_size_read(inode);
1484 const pgoff_t end_index = ((unsigned long long) i_size)
1486 loff_t psize = (page->index + 1) << PAGE_SHIFT;
1487 unsigned offset = 0;
1488 bool need_balance_fs = false;
1490 struct f2fs_io_info fio = {
1494 .op_flags = wbc_to_write_flags(wbc),
1495 .old_blkaddr = NULL_ADDR,
1497 .encrypted_page = NULL,
1499 .need_lock = LOCK_RETRY,
1503 trace_f2fs_writepage(page, DATA);
1505 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1508 if (page->index < end_index)
1512 * If the offset is out-of-range of file size,
1513 * this page does not have to be written to disk.
1515 offset = i_size & (PAGE_SIZE - 1);
1516 if ((page->index >= end_index + 1) || !offset)
1519 zero_user_segment(page, offset, PAGE_SIZE);
1521 if (f2fs_is_drop_cache(inode))
1523 /* we should not write 0'th page having journal header */
1524 if (f2fs_is_volatile_file(inode) && (!page->index ||
1525 (!wbc->for_reclaim &&
1526 available_free_memory(sbi, BASE_CHECK))))
1529 /* we should bypass data pages to proceed the kworkder jobs */
1530 if (unlikely(f2fs_cp_error(sbi))) {
1531 mapping_set_error(page->mapping, -EIO);
1535 /* Dentry blocks are controlled by checkpoint */
1536 if (S_ISDIR(inode->i_mode)) {
1537 fio.need_lock = LOCK_DONE;
1538 err = do_write_data_page(&fio);
1542 if (!wbc->for_reclaim)
1543 need_balance_fs = true;
1544 else if (has_not_enough_free_secs(sbi, 0, 0))
1547 set_inode_flag(inode, FI_HOT_DATA);
1550 if (f2fs_has_inline_data(inode)) {
1551 err = f2fs_write_inline_data(inode, page);
1556 if (err == -EAGAIN) {
1557 err = do_write_data_page(&fio);
1558 if (err == -EAGAIN) {
1559 fio.need_lock = LOCK_REQ;
1560 err = do_write_data_page(&fio);
1563 if (F2FS_I(inode)->last_disk_size < psize)
1564 F2FS_I(inode)->last_disk_size = psize;
1567 if (err && err != -ENOENT)
1571 inode_dec_dirty_pages(inode);
1573 ClearPageUptodate(page);
1575 if (wbc->for_reclaim) {
1576 f2fs_submit_merged_write_cond(sbi, inode, 0, page->index, DATA);
1577 clear_inode_flag(inode, FI_HOT_DATA);
1578 remove_dirty_inode(inode);
1583 if (!S_ISDIR(inode->i_mode))
1584 f2fs_balance_fs(sbi, need_balance_fs);
1586 if (unlikely(f2fs_cp_error(sbi))) {
1587 f2fs_submit_merged_write(sbi, DATA);
1592 *submitted = fio.submitted;
1597 redirty_page_for_writepage(wbc, page);
1599 return AOP_WRITEPAGE_ACTIVATE;
1604 static int f2fs_write_data_page(struct page *page,
1605 struct writeback_control *wbc)
1607 return __write_data_page(page, NULL, wbc, FS_DATA_IO);
1611 * This function was copied from write_cche_pages from mm/page-writeback.c.
1612 * The major change is making write step of cold data page separately from
1613 * warm/hot data page.
1615 static int f2fs_write_cache_pages(struct address_space *mapping,
1616 struct writeback_control *wbc,
1617 enum iostat_type io_type)
1621 struct pagevec pvec;
1623 pgoff_t uninitialized_var(writeback_index);
1625 pgoff_t end; /* Inclusive */
1627 pgoff_t last_idx = ULONG_MAX;
1629 int range_whole = 0;
1632 pagevec_init(&pvec, 0);
1634 if (get_dirty_pages(mapping->host) <=
1635 SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
1636 set_inode_flag(mapping->host, FI_HOT_DATA);
1638 clear_inode_flag(mapping->host, FI_HOT_DATA);
1640 if (wbc->range_cyclic) {
1641 writeback_index = mapping->writeback_index; /* prev offset */
1642 index = writeback_index;
1649 index = wbc->range_start >> PAGE_SHIFT;
1650 end = wbc->range_end >> PAGE_SHIFT;
1651 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1653 cycled = 1; /* ignore range_cyclic tests */
1655 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1656 tag = PAGECACHE_TAG_TOWRITE;
1658 tag = PAGECACHE_TAG_DIRTY;
1660 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1661 tag_pages_for_writeback(mapping, index, end);
1663 while (!done && (index <= end)) {
1666 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
1667 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1);
1671 for (i = 0; i < nr_pages; i++) {
1672 struct page *page = pvec.pages[i];
1673 bool submitted = false;
1675 if (page->index > end) {
1680 done_index = page->index;
1684 if (unlikely(page->mapping != mapping)) {
1690 if (!PageDirty(page)) {
1691 /* someone wrote it for us */
1692 goto continue_unlock;
1695 if (PageWriteback(page)) {
1696 if (wbc->sync_mode != WB_SYNC_NONE)
1697 f2fs_wait_on_page_writeback(page,
1700 goto continue_unlock;
1703 BUG_ON(PageWriteback(page));
1704 if (!clear_page_dirty_for_io(page))
1705 goto continue_unlock;
1707 ret = __write_data_page(page, &submitted, wbc, io_type);
1708 if (unlikely(ret)) {
1710 * keep nr_to_write, since vfs uses this to
1711 * get # of written pages.
1713 if (ret == AOP_WRITEPAGE_ACTIVATE) {
1717 } else if (ret == -EAGAIN) {
1719 if (wbc->sync_mode == WB_SYNC_ALL) {
1721 congestion_wait(BLK_RW_ASYNC,
1727 done_index = page->index + 1;
1730 } else if (submitted) {
1731 last_idx = page->index;
1734 /* give a priority to WB_SYNC threads */
1735 if ((atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) ||
1736 --wbc->nr_to_write <= 0) &&
1737 wbc->sync_mode == WB_SYNC_NONE) {
1742 pagevec_release(&pvec);
1746 if (!cycled && !done) {
1749 end = writeback_index - 1;
1752 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1753 mapping->writeback_index = done_index;
1755 if (last_idx != ULONG_MAX)
1756 f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
1762 int __f2fs_write_data_pages(struct address_space *mapping,
1763 struct writeback_control *wbc,
1764 enum iostat_type io_type)
1766 struct inode *inode = mapping->host;
1767 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1768 struct blk_plug plug;
1771 /* deal with chardevs and other special file */
1772 if (!mapping->a_ops->writepage)
1775 /* skip writing if there is no dirty page in this inode */
1776 if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
1779 /* during POR, we don't need to trigger writepage at all. */
1780 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1783 if (S_ISDIR(inode->i_mode) && wbc->sync_mode == WB_SYNC_NONE &&
1784 get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
1785 available_free_memory(sbi, DIRTY_DENTS))
1788 /* skip writing during file defragment */
1789 if (is_inode_flag_set(inode, FI_DO_DEFRAG))
1792 trace_f2fs_writepages(mapping->host, wbc, DATA);
1794 /* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
1795 if (wbc->sync_mode == WB_SYNC_ALL)
1796 atomic_inc(&sbi->wb_sync_req);
1797 else if (atomic_read(&sbi->wb_sync_req))
1800 blk_start_plug(&plug);
1801 ret = f2fs_write_cache_pages(mapping, wbc, io_type);
1802 blk_finish_plug(&plug);
1804 if (wbc->sync_mode == WB_SYNC_ALL)
1805 atomic_dec(&sbi->wb_sync_req);
1807 * if some pages were truncated, we cannot guarantee its mapping->host
1808 * to detect pending bios.
1811 remove_dirty_inode(inode);
1815 wbc->pages_skipped += get_dirty_pages(inode);
1816 trace_f2fs_writepages(mapping->host, wbc, DATA);
1820 static int f2fs_write_data_pages(struct address_space *mapping,
1821 struct writeback_control *wbc)
1823 struct inode *inode = mapping->host;
1825 return __f2fs_write_data_pages(mapping, wbc,
1826 F2FS_I(inode)->cp_task == current ?
1827 FS_CP_DATA_IO : FS_DATA_IO);
1830 static void f2fs_write_failed(struct address_space *mapping, loff_t to)
1832 struct inode *inode = mapping->host;
1833 loff_t i_size = i_size_read(inode);
1836 down_write(&F2FS_I(inode)->i_mmap_sem);
1837 truncate_pagecache(inode, i_size);
1838 truncate_blocks(inode, i_size, true);
1839 up_write(&F2FS_I(inode)->i_mmap_sem);
1843 static int prepare_write_begin(struct f2fs_sb_info *sbi,
1844 struct page *page, loff_t pos, unsigned len,
1845 block_t *blk_addr, bool *node_changed)
1847 struct inode *inode = page->mapping->host;
1848 pgoff_t index = page->index;
1849 struct dnode_of_data dn;
1851 bool locked = false;
1852 struct extent_info ei = {0,0,0};
1856 * we already allocated all the blocks, so we don't need to get
1857 * the block addresses when there is no need to fill the page.
1859 if (!f2fs_has_inline_data(inode) && len == PAGE_SIZE &&
1860 !is_inode_flag_set(inode, FI_NO_PREALLOC))
1863 if (f2fs_has_inline_data(inode) ||
1864 (pos & PAGE_MASK) >= i_size_read(inode)) {
1865 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, true);
1869 /* check inline_data */
1870 ipage = get_node_page(sbi, inode->i_ino);
1871 if (IS_ERR(ipage)) {
1872 err = PTR_ERR(ipage);
1876 set_new_dnode(&dn, inode, ipage, ipage, 0);
1878 if (f2fs_has_inline_data(inode)) {
1879 if (pos + len <= MAX_INLINE_DATA(inode)) {
1880 read_inline_data(page, ipage);
1881 set_inode_flag(inode, FI_DATA_EXIST);
1883 set_inline_node(ipage);
1885 err = f2fs_convert_inline_page(&dn, page);
1888 if (dn.data_blkaddr == NULL_ADDR)
1889 err = f2fs_get_block(&dn, index);
1891 } else if (locked) {
1892 err = f2fs_get_block(&dn, index);
1894 if (f2fs_lookup_extent_cache(inode, index, &ei)) {
1895 dn.data_blkaddr = ei.blk + index - ei.fofs;
1898 err = get_dnode_of_data(&dn, index, LOOKUP_NODE);
1899 if (err || dn.data_blkaddr == NULL_ADDR) {
1900 f2fs_put_dnode(&dn);
1901 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO,
1909 /* convert_inline_page can make node_changed */
1910 *blk_addr = dn.data_blkaddr;
1911 *node_changed = dn.node_changed;
1913 f2fs_put_dnode(&dn);
1916 __do_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO, false);
1920 static int f2fs_write_begin(struct file *file, struct address_space *mapping,
1921 loff_t pos, unsigned len, unsigned flags,
1922 struct page **pagep, void **fsdata)
1924 struct inode *inode = mapping->host;
1925 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1926 struct page *page = NULL;
1927 pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
1928 bool need_balance = false;
1929 block_t blkaddr = NULL_ADDR;
1932 trace_f2fs_write_begin(inode, pos, len, flags);
1935 * We should check this at this moment to avoid deadlock on inode page
1936 * and #0 page. The locking rule for inline_data conversion should be:
1937 * lock_page(page #0) -> lock_page(inode_page)
1940 err = f2fs_convert_inline_inode(inode);
1946 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
1947 * wait_for_stable_page. Will wait that below with our IO control.
1949 page = pagecache_get_page(mapping, index,
1950 FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
1958 err = prepare_write_begin(sbi, page, pos, len,
1959 &blkaddr, &need_balance);
1963 if (need_balance && has_not_enough_free_secs(sbi, 0, 0)) {
1965 f2fs_balance_fs(sbi, true);
1967 if (page->mapping != mapping) {
1968 /* The page got truncated from under us */
1969 f2fs_put_page(page, 1);
1974 f2fs_wait_on_page_writeback(page, DATA, false);
1976 /* wait for GCed encrypted page writeback */
1977 if (f2fs_encrypted_file(inode))
1978 f2fs_wait_on_encrypted_page_writeback(sbi, blkaddr);
1980 if (len == PAGE_SIZE || PageUptodate(page))
1983 if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode)) {
1984 zero_user_segment(page, len, PAGE_SIZE);
1988 if (blkaddr == NEW_ADDR) {
1989 zero_user_segment(page, 0, PAGE_SIZE);
1990 SetPageUptodate(page);
1994 bio = f2fs_grab_bio(inode, blkaddr, 1);
1999 bio->bi_opf = REQ_OP_READ;
2000 if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
2006 __submit_bio(sbi, bio, DATA);
2009 if (unlikely(page->mapping != mapping)) {
2010 f2fs_put_page(page, 1);
2013 if (unlikely(!PageUptodate(page))) {
2021 f2fs_put_page(page, 1);
2022 f2fs_write_failed(mapping, pos + len);
2026 static int f2fs_write_end(struct file *file,
2027 struct address_space *mapping,
2028 loff_t pos, unsigned len, unsigned copied,
2029 struct page *page, void *fsdata)
2031 struct inode *inode = page->mapping->host;
2033 trace_f2fs_write_end(inode, pos, len, copied);
2036 * This should be come from len == PAGE_SIZE, and we expect copied
2037 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
2038 * let generic_perform_write() try to copy data again through copied=0.
2040 if (!PageUptodate(page)) {
2041 if (unlikely(copied != len))
2044 SetPageUptodate(page);
2049 set_page_dirty(page);
2051 if (pos + copied > i_size_read(inode))
2052 f2fs_i_size_write(inode, pos + copied);
2054 f2fs_put_page(page, 1);
2055 f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
2059 static int check_direct_IO(struct inode *inode, struct iov_iter *iter,
2062 unsigned blocksize_mask = inode->i_sb->s_blocksize - 1;
2064 if (offset & blocksize_mask)
2067 if (iov_iter_alignment(iter) & blocksize_mask)
2073 static ssize_t f2fs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
2075 struct address_space *mapping = iocb->ki_filp->f_mapping;
2076 struct inode *inode = mapping->host;
2077 size_t count = iov_iter_count(iter);
2078 loff_t offset = iocb->ki_pos;
2079 int rw = iov_iter_rw(iter);
2082 err = check_direct_IO(inode, iter, offset);
2086 if (__force_buffered_io(inode, rw))
2089 trace_f2fs_direct_IO_enter(inode, offset, count, rw);
2091 down_read(&F2FS_I(inode)->dio_rwsem[rw]);
2092 err = blockdev_direct_IO(iocb, inode, iter, get_data_block_dio);
2093 up_read(&F2FS_I(inode)->dio_rwsem[rw]);
2097 f2fs_update_iostat(F2FS_I_SB(inode), APP_DIRECT_IO,
2099 set_inode_flag(inode, FI_UPDATE_WRITE);
2100 } else if (err < 0) {
2101 f2fs_write_failed(mapping, offset + count);
2105 trace_f2fs_direct_IO_exit(inode, offset, count, rw, err);
2110 void f2fs_invalidate_page(struct page *page, unsigned int offset,
2111 unsigned int length)
2113 struct inode *inode = page->mapping->host;
2114 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2116 if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
2117 (offset % PAGE_SIZE || length != PAGE_SIZE))
2120 if (PageDirty(page)) {
2121 if (inode->i_ino == F2FS_META_INO(sbi)) {
2122 dec_page_count(sbi, F2FS_DIRTY_META);
2123 } else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
2124 dec_page_count(sbi, F2FS_DIRTY_NODES);
2126 inode_dec_dirty_pages(inode);
2127 remove_dirty_inode(inode);
2131 /* This is atomic written page, keep Private */
2132 if (IS_ATOMIC_WRITTEN_PAGE(page))
2133 return drop_inmem_page(inode, page);
2135 set_page_private(page, 0);
2136 ClearPagePrivate(page);
2139 int f2fs_release_page(struct page *page, gfp_t wait)
2141 /* If this is dirty page, keep PagePrivate */
2142 if (PageDirty(page))
2145 /* This is atomic written page, keep Private */
2146 if (IS_ATOMIC_WRITTEN_PAGE(page))
2149 set_page_private(page, 0);
2150 ClearPagePrivate(page);
2155 * This was copied from __set_page_dirty_buffers which gives higher performance
2156 * in very high speed storages. (e.g., pmem)
2158 void f2fs_set_page_dirty_nobuffers(struct page *page)
2160 struct address_space *mapping = page->mapping;
2161 unsigned long flags;
2163 if (unlikely(!mapping))
2166 spin_lock(&mapping->private_lock);
2167 lock_page_memcg(page);
2169 spin_unlock(&mapping->private_lock);
2171 spin_lock_irqsave(&mapping->tree_lock, flags);
2172 WARN_ON_ONCE(!PageUptodate(page));
2173 account_page_dirtied(page, mapping);
2174 radix_tree_tag_set(&mapping->page_tree,
2175 page_index(page), PAGECACHE_TAG_DIRTY);
2176 spin_unlock_irqrestore(&mapping->tree_lock, flags);
2177 unlock_page_memcg(page);
2179 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
2183 static int f2fs_set_data_page_dirty(struct page *page)
2185 struct address_space *mapping = page->mapping;
2186 struct inode *inode = mapping->host;
2188 trace_f2fs_set_page_dirty(page, DATA);
2190 if (!PageUptodate(page))
2191 SetPageUptodate(page);
2193 if (f2fs_is_atomic_file(inode) && !f2fs_is_commit_atomic_write(inode)) {
2194 if (!IS_ATOMIC_WRITTEN_PAGE(page)) {
2195 register_inmem_page(inode, page);
2199 * Previously, this page has been registered, we just
2205 if (!PageDirty(page)) {
2206 f2fs_set_page_dirty_nobuffers(page);
2207 update_dirty_page(inode, page);
2213 static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
2215 struct inode *inode = mapping->host;
2217 if (f2fs_has_inline_data(inode))
2220 /* make sure allocating whole blocks */
2221 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2222 filemap_write_and_wait(mapping);
2224 return generic_block_bmap(mapping, block, get_data_block_bmap);
2227 #ifdef CONFIG_MIGRATION
2228 #include <linux/migrate.h>
2230 int f2fs_migrate_page(struct address_space *mapping,
2231 struct page *newpage, struct page *page, enum migrate_mode mode)
2233 int rc, extra_count;
2234 struct f2fs_inode_info *fi = F2FS_I(mapping->host);
2235 bool atomic_written = IS_ATOMIC_WRITTEN_PAGE(page);
2237 BUG_ON(PageWriteback(page));
2239 /* migrating an atomic written page is safe with the inmem_lock hold */
2240 if (atomic_written) {
2241 if (mode != MIGRATE_SYNC)
2243 if (!mutex_trylock(&fi->inmem_lock))
2248 * A reference is expected if PagePrivate set when move mapping,
2249 * however F2FS breaks this for maintaining dirty page counts when
2250 * truncating pages. So here adjusting the 'extra_count' make it work.
2252 extra_count = (atomic_written ? 1 : 0) - page_has_private(page);
2253 rc = migrate_page_move_mapping(mapping, newpage,
2254 page, NULL, mode, extra_count);
2255 if (rc != MIGRATEPAGE_SUCCESS) {
2257 mutex_unlock(&fi->inmem_lock);
2261 if (atomic_written) {
2262 struct inmem_pages *cur;
2263 list_for_each_entry(cur, &fi->inmem_pages, list)
2264 if (cur->page == page) {
2265 cur->page = newpage;
2268 mutex_unlock(&fi->inmem_lock);
2273 if (PagePrivate(page))
2274 SetPagePrivate(newpage);
2275 set_page_private(newpage, page_private(page));
2277 migrate_page_copy(newpage, page);
2279 return MIGRATEPAGE_SUCCESS;
2283 const struct address_space_operations f2fs_dblock_aops = {
2284 .readpage = f2fs_read_data_page,
2285 .readpages = f2fs_read_data_pages,
2286 .writepage = f2fs_write_data_page,
2287 .writepages = f2fs_write_data_pages,
2288 .write_begin = f2fs_write_begin,
2289 .write_end = f2fs_write_end,
2290 .set_page_dirty = f2fs_set_data_page_dirty,
2291 .invalidatepage = f2fs_invalidate_page,
2292 .releasepage = f2fs_release_page,
2293 .direct_IO = f2fs_direct_IO,
2295 #ifdef CONFIG_MIGRATION
2296 .migratepage = f2fs_migrate_page,